One reason is fairness. Some papers are rejected because they just aren't good enough, but others could plausibly have been accepted if the deliberations had by chance gone a little differently. When a reasonable submission is rejected due to bad luck, it feels like adding insult to injury if the author then has the subsidize the registration fees for the accepted papers. (It might be reasonable to expect authors to cover the cost of reviewing the submissions, but that's far less than typical registration fees.)
Another reason is to minimize bureaucratic overhead. Whenever you collect fees, you have to negotiate with authors over who deserves a fee waiver, and you have to deal with unusual situations (authors who can't or won't pay online, chargebacks from disputed transactions, etc.). The pain scales with the number of people involved, and it tends to be extra high if you try anything nonstandard, so there's an incentive not to experiment, especially in ways that increase the number of transactions.
Another consideration is keeping authors happy. In any review process, some authors are going to end up disgruntled. Maybe they actually received low-quality reviews, or maybe it just feels that way to them, but either way they may complain. I'd bet that complaints would increase tenfold if the disgruntled authors had been charged $200 and felt entitled to a corresponding level of service. I certainly wouldn't want to be in charge of dealing with the complaints.